VoIP Reliability White Paper 102604
October 2004
Voice over IP
Reliability
Architecture Matters
Comparing Three Different Approaches to Delivering 99.999% Availability
Phone service is like air in today’s businesses: Its constant availability is both required
and assumed. This state of affairs has arisen from more than a century of efforts by the
makers of traditional voice switches to develop systems that are available at least
99.999% of the time. Concerns that voice over IP (VoIP) may not offer this vaunted “five
nines” availability has been one of the major impediments to convergence.
This is ironic, because achieving a much higher level of communications-network
resilience was the fundamental design goal for what is now the Internet. In contrast to
their circuit-switched counterparts, packet-switched networks fundamentally enable fault
tolerance, adaptive routing, and disaster recovery. It is quite possible—and can be quite
cost-effective—to build IP-based voice systems that are more reliable than circuit-
switched PBX platforms. The key is to start with the right foundation.
Today’s VoIP solutions fall into three basic categories: Systems evolved from traditional
PBX platforms, systems evolved from traditional data-switch platforms, and systems
designed from the ground up for VoIP. All three of these architectures can be used to
deliver VoIP systems with five-nines reliability, but they involve different degrees of
complexity and cost. In this paper, we will examine the effects that these different
architectures have upon the ability to deliver IP-based voice systems that are both highly
available and cost effective.
The Legacy Voice Approach to VoIP Reliability
The manufacturers of legacy voice systems have a great history of delivering extremely
reliable voice switches, and in fact are responsible for the 99.999% availability standard
that VoIP solutions must match or exceed. However, legacy PBXes and key systems are
hierarchical voice silos that operate independently at each location in a multi-site
company. They cannot back each other up or be managed as a single voice network, and
create a single point of failure at each site.
The VoIP solutions that have been evolved out of these legacy voice switches have
inherited some of the inherent fragmentation of this centralized and hierarchical
architecture.
VoIP Reliability 10/26/2004 2 of 6
One approach is to put a centralized IP PBX at the main site provisioning dial tone over
an IP backbone to IP phones at the remote locations. If the WAN link goes down, the IP
phones become useless, so availability assurance can be increased by installing a small
standby IP PBX at each remote site. This device might be able to sense an outage and
automatically take over as necessary, or it might require someone to flip a manual switch
after employees start complaining that there is no dial tone, and then flip the switch back
after the WAN link is restored. The cost of the failover solution is generally inversely
proportional to its level of automation. If a company has multiple sites that are fairly
large, separate IP PBXes are installed at each location.
The result is switches that operate more like separate silos than elements of a single voice
system. This fragmented architecture cripples one of the key benefits of VoIP: the ability
to create a single voice system that is distributed across multiple sites and can cover for
individual switches that fail. It can also increase costs. Each of the silos may still require
outsourced management and maintenance by a local teleconnect, just as their circuit-
switched predecessors did. There may also be licensing issues to consider.
The Data-Centric Approach to VoIP Reliability
The term “convergence” sounds like a merger of roughly equal parties, but in the
voice/data arena it is actually more of an acquisition. Voice is being turned into another
network application, albeit a very important one. Given this fundamental truth, it is
tempting to assume that the established data-switch vendors have an architectural edge as
they develop VoIP platforms. However, the data world carries some legacy baggage of its
own.
To begin with, while voice does not require much bandwidth, each conversation has to be
maintained in a constant stream with strict latency thresholds for acceptable voice
quality. Data traffic is very forgiving of latency, and the switches were designed to burst
massive amounts of data through as quickly as possible.
The data-switch vendors approach VoIP by taking these basic, data-optimized Ethernet
switch platforms and embedding some technology into them. Call control is implemented
in a separate centralized server located in a data center. This centralized server is
fundamental to the set up and tear down of all calls to, from, or between IP phones at all
the company’s different sites, so it creates a single point of failure. Multiple call-control
servers can be purchased at additional expense and clustered together to provide fault
tolerance, but this centralized architecture still assumes that the IP backbone connecting
the various offices is always available.
VoIP Reliability 10/26/2004 3 of 6
When the WAN does go down, the remote offices can fall back to a survival mode—if
this optional feature was purchased and installed, and the IP phones were configured to
use it in the event of a WAN outage. The approach is similar to that of the legacy voice
vendors, who increase reliability by adding a standby PBX at the remote site. For the
legacy data vendors, the enhanced availability takes the shape of a card that is installed in
the remote site’s data switch. This survival-mode feature may not be included in the
quoted price of the VoIP solution, and it delivers a reduced level of voice service: Users
limp along with dial tone and a few basic features.
When VoIP is based on a data-switch architecture, availability is increased by building in
a lot of redundancies. This over-provisioning approach can get very expensive, and it also
makes voice systems more complex. In fact, the inherent complexity of retrofitting a data
platform for voice adds reliability challenges at every level: design, implementation, day-
to-day operations, and problem resolution. Implementing a VoIP solution can involve
sifting through hundreds of devices and options and trying to figure out which ones must
be cobbled together to provide basic VoIP functionality with reasonable availability.
Configuration of survival-mode features can require more than 20 command-line entries,
each key peck adding one more opportunity for human error.
The VoIP-by-Design Advantage
IP networks are inherently distributed and resilient, and VoIP architects starting with a
blank slate can exploit this fundamental strength to create a self-healing voice platform.
A truly unified voice system can be distributed across multiple sites by using a simple
peer-to-peer architecture that has no single point of failure. IP voice switches designed
specifically for voice can each incorporate a complete call processor — even small
models aimed at remote offices requiring eight ports or less. Each switch is a peer with a
full complement of routing information safely held in local flash memory, and can
operate as a standalone PBX if its site is cut off from the IP backbone. It can make best-
effort calls on its own, using a failover PSTN trunk if necessary. When switches are
added or restored to the network, they and the existing switches at all the sites
automatically discover each other and start working together.
VoIP Reliability 10/26/2004 4 of 6
If a switch providing PSTN access to one site were to fail, its peer switches elsewhere in
the WAN would provide alternate PSTN access to the users at that site. They would
continue to get a full set of voice features, not a survival-mode subset. As long as the data
backbone stays up, this type of distributed voice network can’t have an outage unless all
the switches go down simultaneously.
Reliability thus comes built in with this approach, and the five-nines availability
requirement for voice is easily met. In fact, it can be increased to ten nines by installing a
redundant switch with PSTN access at each site. A native VoIP architecture thus has
built-in autonomy and survivability. Delivering a system that is reliable and highly
available doesn’t have to involve additional cost or complexity.
Conclusion
When disruptive technologies start to emerge, established vendors always look at the new
landscape through the lens of the world they currently dominate. The architectures they
come up with are limited or even crippled by this legacy. We all know that computers,
generally speaking, have not been as reliable as phones, and there is still a lot of FUD
(Fear, Uncertainty, and Doubt) circulating about VoIP. And in fact, there have been VoIP
implementations that cannot be called reliable.
However, there is no longer any question that highly available VoIP systems can be
delivered on any of the three basic architectures discussed in this paper. It is now a
question of how such reliability is achieved, not whether it can be achieved. For VoIP
solutions built on what is essentially a legacy voice or legacy data foundation, each
additional nine in the availability rating comes at the expense of a lot more complexity
and resources. For a voice system that is VoIP by design—mimicking the resilience of IP
networks with a peer-to-peer communications architecture, and with autonomous call
management built into each voice switch—it comes very naturally.
VoIP Reliability 10/26/2004 5 of 6
960 Stewart Drive
Sunnyvale, CA 94085
(408) 331-3300
1-800-425-9385
Fax: (408) 331-3333
Email: [email protected]
www.shoretel.com
© 2004 ShoreTel, Inc. All
rights reserved. October 2004
